Assessing the risk of a historical-level flood is essential for regional flood protection and resilience establishment. However, due to the limited spatiotemporal coverage of observations, the impact assessment relies on model simulations and is thus subject to uncertainties from cascade physical processes. This study assesses the flood hazard map with uncertainties subject to different combinations of runoff inputs, variables for flood frequency analysis and fitting distributions based on estimations by the CaMa-Flood global hydrodynamic model. Our results show that deviation in the runoff inputs is the most influential source of uncertainties in the estimated flooded water depth and inundation area, contributing more than 80 % of the total uncertainties investigated in this study. Global and regional inundation maps for floods with 1-in-100 year return periods show large uncertainty values but small uncertainty ratios for river channels and lakes, while the opposite results are found for dry zones and mountainous regions. This uncertainty is a result of increasing variation at tails among various fitting distributions. In addition, the uncertainty between selected variables is limited but increases from the regular period to the rarer floods, both for the water depth at points and for inundation area over regions. The uncertainties in inundation area also lead to uncertainties in estimating the population and economy exposure to the floods. In total, inundation accounts for 9.1 % [8.1 %–10.3 %] of the land area for a 1-in-100 year flood, leading to 13.4 % [12.1 %–15 %] of population exposure and 13.1 % [11.8 %–14.7 %] of economic exposure for the globe. The flood exposure and uncertainties vary by continent and the results in Africa have the largest uncertainty, probably due to the limited observations to constrain runoff simulations, indicating a necessity to improve the performance of different hydrological models especially for data-limited regions.

The impact of human water management on river discharge is increasingly viewed as a missing process in Earth system modeling. Models which have attempted to include it are generally at coarse resolution and uncoupled to the atmosphere. We propose to describe human water management at high spatial resolution using ORganizing Carbon and Hydrology In Dynamic EcosystEms concept of hydrological‐transfer‐units in the routing parametrization. Irrigated areas are linked to river abstraction points using a minimization process. The directed graphs of river flows and adduction network for irrigation is transposed to propagate water demands upstream. Reservoirs and dams are placed along the graph to balance supply and demands in the four chosen water value classes. Dam regulation is assumed to maximize demand satisfaction and dampen floods while respecting the properties of the infrastructure. The developed human water management module is applied to the Yellow River where irrigation and dam regulation are known to have a strong impact. Results show that the impact of human water management is strongly heterogeneous over space. It propagates along the river channels and can be mitigated by the confluence of tributaries. Moreover, the human impact has a strong seasonality due to time varying irrigation demands and the response of dam regulation. A number of uncertainties still remain and affect the simulated river discharge. Nevertheless, the representation of human water management improves the model's behavior in terms of magnitude and intra‐annual variations of river discharge, and offers the opportunity to implement anthropogenic processes in the water cycle of Earth System Models.

In China, irrigation is widespread in 40.7% cropland to sustain crop yields. By its action on water cycle, irrigation affects water resources, and local climate. In this study, a new irrigation module, including flood and paddy irrigation technologies, was developed in the ORCHIDEE‐CROP land surface model which describes crop phenology and growth in order to estimate irrigation demands over China from 1982 to 2014. Three simulations were performed including NI: no irrigation; IR: with irrigation limited by local water resources; and FI: with irrigation demand fulfilled. Observations and census data were used to validate the simulations. Results showed that the estimated irrigation water withdrawal (W) based on IR and FI scenarios bracket statistical W with fair spatial agreements (r = 0.68 ± 0.07; p < 0.01). Improving irrigation efficiency was found to be the dominant factor leading to the observed W decrease. By comparing simulated total water storage (TWS) with GRACE observations, we found that simulated TWS with irrigation well explained the TWS variation over China. However, our simulation overestimated the seasonality of TWS in the Yangtze River Basin due to ignoring regulation of artificial reservoirs. The observed TWS decrease in the Yellow River Basin caused by groundwater depletion was not totally captured in our simulation, but it can be inferred by combining simulated TWS with census data. Moreover, we demonstrated that land use change tended to drive W locally, but had little effect on total W over China due to water resources limitation.